OpenCode Turns GitHub Triage Into Team Memory

The compelling shift is from static prompt engineering to accumulated operational memory: the harness gradually learns which specialist fits each failure mode. The tradeoff is that durable context and delegation need strong observability and permission boundaries.

• –Recurring issue patterns can produce specialized subagents without upfront manual configuration
• –Long-lived sessions reduce repeated explanations and preserve project-specific debugging context
• –GitHub triage is a strong proving ground because classification, ownership, reproduction, and fixes require different expertise
• –Parallel delegation can improve throughput but increases token costs and makes failures harder to trace
• –Worktree isolation and strict permissions remain essential when agents act on untrusted issue text